Share on LinkedIn

Applying machine learning in the search for life-saving cancer drugs

Maxine-Laurie Marshall — April 2017
Machine learning is revolutionizing cancer drug discovery. Dr Bissan Al-Lazikani, head of data science at the Institute of Cancer Research, explains how the technology is adding a new dimension to the fight against the disease.

Spurred on by surging volumes of data and powerful, inexpensive computing capacity, machine learning is now being applied to a host of previously ‘impossible challenges’ — from self-driving cars and marketing personalization to fraud detection and simultaneous translation.

But there can hardly be a more valuable field for its application than in the battle against cancer. A computational approach to building models using algorithms that iteratively learn from the data they are exposed to — without being explicitly programmed — is already having a revolutionary impact on the discovery of effective new drugs and therapies.

This technological leap forward is progressively supplanting the historical approach of scientists, who would normally float a hypothesis about a drug’s potential and then observe its effects in experimentation and trials. Machine learning is now showing researchers the areas that are most likely to be successful targets for new drugs, reducing the need for many fruitless trials.

Dr Bissan Al-Lazikani, head of data science at the Institute of Cancer Research (ICR), highlights how machine learning is fuelled by data — and lots of it. This is something ICR has in abundance — including genomes, imaging, treatment and healthcare data of cancer and healthy patients.

At ICR, these rich sources are brought together in a single platform — the Knowledge Hub — that allows researchers to search for patterns and explore the complexities of specific cancers at an unprecedented level. With machine-learning techniques, the team is  building models to predict how tumours will respond to treatment and which drug is best suited to treating the disease.

For any one cancer, there might be 100 possible targets or genes that a drug can be used against. That process of drug discovery used to be somewhat hit-or-miss, says Al-Lazikani, so by the time a researcher realized they had chosen a wrong target, a lot of time and effort had been wasted. With machine learning, the data is viewed in a smarter way that helps scientists to ‘hedge their bets.’

“We have developed machine-learning techniques and algorithms to analyze the comprehensive integrated data and come up with suggestions for the best targets to work on given a specific context. The algorithms learn from the examples they’ve seen before. So when they encounter a protein that is new to them they can tell us whether there is a chance that it might make a good drug target or whether it is likely to be a failure. Drug discovery at the ICR now uses machine learning to help pick the targets we are going to work on,” she says.

There are many benefits: patients will receive smarter, more effective treatment and researchers can work far more efficiently, often coming across valuable information that they may not have known even existed. “It’s hidden knowledge discovery that is the really powerful result of machine learning,” Al-Lazikani sums up.
Network analytics applied to biology

The data that fuels machine-learning algorithms can come in some surprising forms. “You are no longer restricted to biology, you can apply any kind of analytics to it,” says Al- Lazikani. For example, researchers can apply network analytic models to help uncover patterns of communication and interaction across the thousands of genes and proteins in the body.

As Al-Lazikani highlights, the associations of behavior of proteins and genes can resemble how friendship networks are built on social networks such as Facebook or LinkedIn.

“Our study is the first to identify the rules of ‘social behaviour’ of cancer proteins and use it to predict new targets for potential cancer drugs. We have calculated almost 400 different network-based properties that have nothing to do with biology. By feeding the data into different machine-learning algorithms we can come up with predictive models.”

As she outlines, the approach has produced some strong results: “We’ve found that cancer proteins and cancer drug targets actually behave very differently to other proteins in the body. That means we can build predictive models that look at how a protein behaves in a cancer cell and predict whether it’s likely to be a good drug target or not.”

  • Photography: Ben Gold
First published
April 2017
Share on LinkedIn
Dr Bissan Al-Lazikani profile picture
About: Dr Bissan Al-Lazikani
As head of data science at the ICR, Dr Bissan Al-Lazikani brings big data analytics and machine learning into the fight against cancer. At ICR, she has led the development of the world’s largest cancer drug discovery knowledgebase, canSAR. In her spare time she is training for a private pilot’s license.
This website uses cookies to improve functionality. Continue to use the site as normal if you're happy with this or click here to find out more about cookies.